Designing ecosystems of intelligence from first principles

Collective Intelligence, Journal Year: 2024, Volume and Issue: 3(1)

Published: Jan. 1, 2024

This white paper lays out a vision of research and development in the field artificial intelligence for next decade (and beyond). Its denouement is cyber-physical ecosystem natural synthetic sense-making, which humans are integral participants—what we call “shared intelligence.” premised on active inference, formulation adaptive behavior that can be read as physics intelligence, inherits from self-organization. In this context, understand capacity to accumulate evidence generative model one’s sensed world—also known self-evidencing. Formally, corresponds maximizing (Bayesian) evidence, via belief updating over several scales, is, learning, selection. Operationally, self-evidencing realized (variational) message passing or propagation factor graph. Crucially, inference foregrounds an existential imperative intelligent systems; namely, curiosity resolution uncertainty. same underwrites sharing ensembles agents, certain aspects (i.e., factors) each agent’s world provide common ground frame reference. Active plays foundational role ecology sharing—leading formal account collective rests shared narratives goals. We also consider kinds communication protocols must developed enable such intelligences motivate hyper-spatial modeling language transaction protocol, first—and key—step towards ecology.

Language: Английский

---

Resolving uncertainty on the fly: modeling adaptive driving behavior as active inference

Frontiers in Neurorobotics, Journal Year: 2024, Volume and Issue: 18

Published: March 21, 2024

Understanding adaptive human driving behavior, in particular how drivers manage uncertainty, is of key importance for developing simulated driver models that can be used the evaluation and development autonomous vehicles. However, existing traffic psychology behavior either lack computational rigor or only address specific scenarios and/or behavioral phenomena. While developed fields machine learning robotics effectively learn from data, due to their black box nature, they offer little no explanation mechanisms underlying behavior. Thus, generalizable, interpretable, are still rare. This paper proposes such a model based on active inference, modeling framework originating neuroscience. The offers principled solution humans trade progress against caution through policy selection single mandate minimize expected free energy. casts goal-seeking information-seeking (uncertainty-resolving) under objective function, allowing seamlessly resolve uncertainty as means obtain its goals. We apply two apparently disparate require managing (1) past an occluding object (2) visual time-sharing between secondary task, show human-like emerges principle energy minimization.

Language: Английский

---

Supervised structure learning

Biological Psychology, Journal Year: 2024, Volume and Issue: 193, P. 108891 - 108891

Published: Oct. 19, 2024

Language: Английский

---

Deep kinematic inference affords efficient and scalable control of bodily movements

Proceedings of the National Academy of Sciences, Journal Year: 2023, Volume and Issue: 120(51)

Published: Dec. 12, 2023

Performing goal-directed movements requires mapping goals from extrinsic (workspace-relative) to intrinsic (body-relative) coordinates and then motor signals. Mainstream approaches based on optimal control realize the mappings by minimizing cost functions, which is computationally demanding. Instead, active inference uses generative models produce sensory predictions, allows a cheaper inversion However, devising complex kinematic chains like human body challenging. We introduce an architecture that affords simple but effective via easily scales up drive chains. Rich can be specified in both using attractive or repulsive forces. The proposed model reproduces sophisticated bodily paves way for efficient biologically plausible of actuated systems.

Language: Английский

---

The Multiscale Principle in Nature (Principium luxuriæ): Linking Multiscale Thermodynamics to Living and Non-Living Complex Systems

Fractal and Fractional, Journal Year: 2024, Volume and Issue: 8(1), P. 35 - 35

Published: Jan. 4, 2024

Why do fractals appear in so many domains of science? What is the physical principle that generates them? While it true naturally systems, has far been impossible to derive them from first principles. However, a proposed interpretation could shed light on inherent behind creation fractals. This multiscale thermodynamic perspective, which states an increase external energy initiate transport mechanisms facilitate dissipation or release excess at different scales. Within this framework, revealed power law patterns, and lesser extent, fractals, can emerge as geometric manifestation dissipate response forces. In context, exponent these patterns (thermodynamic fractal dimension D) serves indicator balance between entropy production small large Thus, when system more efficient releasing microscopic (macroscopic) level, D tends (decrease). principle, known Principium luxuriæ, may sound promising for describing both complex there still uncertainty about its applicability. work explores physical, astrophysical, sociological, biological systems attempt describe interpret through lens luxuriæ. The analyzed correspond emergent behaviors, chaos theory, turbulence. To cosmic evolution universe geomorphology are examined. Biological such geometry human organs, aging, brain development cognition, moral evolution, Natural Selection, death also analyzed. It found be reinterpreted described dimension. Therefore, defined “Systems interact with each other trigger responses multiple scales manner comes interaction”. That why framework potential uncover new discoveries various fields. For example, suggested reduction generate behavior proliferation complexity numerous fields reinterpretation Selection.

Language: Английский

---

On efficient computation in active inference

Expert Systems with Applications, Journal Year: 2024, Volume and Issue: 253, P. 124315 - 124315

Published: May 31, 2024

Biological agents demonstrate a remarkable proficiency in calibrating appropriate scales of planning and evaluation when interacting with their environments. It follows logically that any decision-making algorithm aspiring to neurobiological plausibility must mirror these attributes, particularly regarding computational expenditure the intricacy evaluative processes. However, active inference encounters notable challenges simulating apt behaviours within complex These stem chiefly from its substantial demands intricate task defining agent's behaviour preference. We address through two-fold approach. First, we introduce by using Bellman-optimality principle minimise cost function (i.e., expected free energy). Briefly, recursively compute energy actions reverse temporal sequence significantly reduce complexity. Secondly, inspired Z-learning algorithm, propose novel method learn time-constrained agent preferences. face-validate efficacy grid-world simulations precise model learning planning, even under uncertainty. algorithmic advances create new opportunities for various applications—in neuroscience machine learning.

Language: Английский

---

An Alternative to Cognitivism: Computational Phenomenology for Deep Learning

Minds and Machines, Journal Year: 2023, Volume and Issue: 33(3), P. 397 - 427

Published: June 29, 2023

Abstract We propose a non-representationalist framework for deep learning relying on novel method computational phenomenology, dialogue between the first-person perspective (relying phenomenology) and mechanisms of models. thereby an alternative to modern cognitivist interpretation learning, according which artificial neural networks encode representations external entities. This mainly relies neuro-representationalism, position that combines strong ontological commitment towards scientific theoretical entities idea brain operates symbolic these proceed as follows: after offering review cognitivism neuro-representationalism in field we first elaborate phenomenological critique positions; then sketch out phenomenology distinguish it from existing alternatives; finally apply this new models trained specific tasks, order formulate conceptual deep-learning, allows one think networks’ terms lived experience.

Language: Английский

---

Disentangling Shape and Pose for Object-Centric Deep Active Inference Models

Communications in computer and information science, Journal Year: 2023, Volume and Issue: unknown, P. 32 - 49

Published: Jan. 1, 2023

Language: Английский

---

Free energy and inference in living systems

Interface Focus, Journal Year: 2023, Volume and Issue: 13(3)

Published: April 14, 2023

Organisms are non-equilibrium, stationary systems self-organized via spontaneous symmetry breaking and undergoing metabolic cycles with broken detailed balance in the environment. The thermodynamic free-energy (FE) principle describes an organism’s homeostasis as regulation of biochemical work constrained by physical FE cost. By contrast, recent research neuroscience theoretical biology explains a higher allostasis Bayesian inference facilitated informational FE. As integrated approach to living systems, this study presents minimization theory overarching essential features both neuroscientific principles. Our results reveal that perception action animals result from active entailed brain, brain operates Schrödinger’s machine conducting neural mechanics minimizing sensory uncertainty. A parsimonious model suggests develops optimal trajectories manifolds induces dynamic bifurcation between attractors process inference.

Language: Английский

---

(HTBNet)Arbitrary Shape Scene Text Detection with Binarization of Hyperbolic Tangent<strong> </strong>and Cross Entropy

Published: May 15, 2024

The existing segmentation-based scene text detection methods mostly need complicated post-processing, and the post-processing operation is separated from training process, which greatly reduces performance. previous method, DBNet successfully simplified integrated into a segmentation network. However, process of model took long time for 1200 epochs sensitivity to texts various scales was lacking, leading some instances being missed. Considering above two problems, we design Network with Binarization Hyperbolic Tangent(HTBNet). First all, propose Tangent (HTB), optimized along which, network can expedite initial convergent speed by reducing amount 600. Because features different channels in same scale feature map focus on information regions image, better represent important all objects devise Multi-Scale Channel Attention(MSCA). Meanwhile considering that multi-scale image cannot be simultaneously detected, novel module named Fused Module Spatial(FMCS), fuse maps channel spatial dimension. Finally adopt cross entropy as loss function, measures difference between predicted values ground truths. experimental results show HTBNet compared lightweight models has achieved competitive performance Total-Text(F-measure:86.0%, FPS:30) MSRA-TD500 (F-measure:87.5%, FPS:30).

Language: Английский

---